Table Of Content∗∗∗∗
Verbs, nouns and affixation
Artemis Alexiadou and Jane Grimshaw
Universität Stuttgart, Rutgers University
What explains the rich patterns of deverbal nominalization? Why do some nouns have
argument structure, while others do not? We seek a solution in which properties of
deverbal nouns are composed from properties of verbs, properties of nouns, and
properties of the morphemes that relate them. The theory of each plus the theory of how
they combine, should give the explanation.
In exploring this, we investigate properties of two theories of nominalization. In
one, the verb-like properties of deverbal nouns result from verbal syntactic structure (a
“structural model”). See, for example, van Hout & Roeper 1998, Fu, Roeper and Borer
1993, 2001, to appear, Alexiadou 2001, to appear). According to the structural
hypothesis, some nouns contain VPs and/or verbal functional layers.
In the other theory, the verbal properties of deverbal nouns result from the event
structure and argument structure of the DPs that they head. By “event structure” we mean
a representation of the elements and structure of a linguistic event, not a representation of
the world. We refer to this view as the “event model”. According to the event model
hypothesis, all derived nouns are represented with the same syntactic structure, the
difference lying in argument structure – which in turn is critically related to event
structure, in the way sketched in Grimshaw (1990), Siloni (1997) among others.1
In pursuing these lines of analysis, and at least to some extent disentangling their
properties, we reach the conclusion that, with respect to a core set of phenomena, the two
theories are remarkably similar – specifically, they achieve success with the same
problems, and must resort to the same stipulations to address the remaining issues that we
discuss (although the stipulations are couched in different forms).
1. Nouns and argument structure: Basic patterns
As shown by Grimshaw (1990), de-verbal nouns do not form a homogeneous
class. Some of them license argument structure and some do not. We do not repeat
the details of the arguments here, but summarise the general points. The central
point for present purposes is that arguments are required only by deverbal nouns
with complex event interpretations. This property, like further differences
between complex event nominals and other deverbal nouns, is obscured by the
fact that many nouns are ambiguous, and can even have three interpretations,
showing different behaviour under each. As (1) illustrates, nouns such as
∗
This paper blends the presentations given by Alexiadou and Grimshaw at the Conference on
Deverbal Nouns held at the University of Lille in 2004. We thank the audience there for lively
commentary and considerable assistance. Since the original talks overlapped in focus, but explored
different stances on the issue of how nominalization is to be understood, the authors decided to
embark on a direct comparison of the two approaches. Both are surprised by the results.
1 We simplify here by treating event structure as a property of verbs, rather than as a property of
verbs in combination with their complements.
Working Papers of the SFB 732 Incremental Specification in Context 01 (2008): 1-16
Schäfer, Florian (ed.)
© 2008 Artemis Alexiadou & Jane Grimshaw
Alexiadou & Grimshaw
examination are three way ambiguous. They can have a “complex event reading”
as in (1a), a “simple event reading” as in (1b), in which they denote an event but
are not associated with an event structure and hence not with an argument
structure, and a further reading in which they refer to the result of an event (1c) or
a participant in it. We will group these together as “individual-referring” or
“individual” nouns.
(1) a. The examination of the patients took a long time (Complex)
b. The examination took a long time (Simple)
c. The examination was on the table (Result)
Only complex event nominals behave like verbs in licensing event-related
PPs, like in an hour, for an hour, see Grimshaw (1990), Zucchi (1993). In this
they behave like verbs. As a consequence of their event structure, they have
argument structure, again like verbs, with the result that they have arguments
which are obligatorily present. They also disallow indefinite determiners (*an
examination of the patients). The examples in (2) are further instances of complex
event nominals, this time formed with –ment. We illustrate properties of complex
event nominals through nouns formed with –ment and –(a)tion throughout the
paper.
(2) a. The frequent payment of your bills keeps your credit rating good.
b. We demanded the replacement of the broken cups in no more than
three days.
Like examination, the nouns in (2) can be individual nominals in addition to their
complex event readings; for example a payment, the replacements are
grammatical. Simple event nominals neither license event-related PPs, nor have
argument structure. Like complex event nominals, simple event nominals denote
events, but syntactically they are similar to individual nominals. Two examples
are event itself, and race in (3b):
(3) a. The event was well organized.
b. The race lasted one hour.
Result nominals and participant nominals likewise fail to license event-
related PPs, lack argument structure, and have no event interpretation at all. Apart
from the fact that they are derived from verbs, individual nouns have the same
syntax as non-verb-related nouns: dog, house, event, trip.
Nominals derived from verbs with no (overt) affix behave as simple event
nouns and/or individual nouns.2 This is true also for irregulars like gift.
2 Some ∅-derived nominals do seem to license arguments (cf. i). The systematicity of these
examples remains to be examined, see Newmeyer (to appear) for discussion.
2
Verbs, nouns and affixation
(4) a. *The constant offer of credit cards to students………….
b. *(The) frequent report of looting …………………
The special behaviour of bare nominals is analyzed in Smith (1972). She
discusses verbs of English which engage in the causative/inchoative alternation,
and nominalize without (overt) affixation. Smith points out that these verbs never
nominalize as “transitive” nouns, but only as nouns with a possessor alone (see
also Chomsky 1970). Examples include change, end and stop, which form
nominals, but not transitive ones. The generalization is visible in these contrasts:
the climate’s change/*global warming’s change of the climate; the race’s
end/*the judge’s end of the race; The train’s unscheduled stop/*The guard’s
unscheduled stop of the train. In fact, using the criteria of Grimshaw (1990), it is
possible to show that these “intransitive” nominals are not complex event
nominals. Their limited interpretations support the claim that zero-derived
nominalization never preserves event structure. Smith also shows that causative
verbs which nominalize with certain affixes show contrasting behaviour: they do
nominalize transitively. We return to this point in Section 6.1.
If we collapse all the nominals discussed so far, and examine them all
together, it appears that nouns can show just about any set of properties.
Grimshaw’s (1990) conclusion was different: that there is a rigid distinction
between nouns which have argument structure and those that do not, which is
obscured by the rather systematic ambiguities illustrated above.
Table 1 compares result nominals with complex event nominals in these and
other respects.
Table 1: Some differences between result and complex event nominals
Result-Nominals Complex Event-Nominals
a. Non-θ-assigner, θ-assigners,
No obligatory arguments Obligatory arguments
b. No event reading Event reading
c. No agent-oriented modifiers Agent-oriented modifiers
d. Subjects are possessives Subjects are arguments
e. by phrases are non-arguments by phrases are arguments
f. No implicit argument control Implicit argument control
g. No aspectual modifiers Aspectual modifiers
h. Modifiers like frequent, constant Modifiers like frequent, constant
only with plural appear with singular
i. May be plural Must be singular
(i) a. My constant change of mentors
b. The frequent release of the prisoners by the governor (David Embick (p.c.)
3
Alexiadou & Grimshaw
2. The generalizations to be explained
Following up on the empirical observations of Section 1, we highlight a core set
of generalizations which must be explained.
1) Only nouns which are related to corresponding verbs have argument
structure. From this we conclude that being associated with an event
structure/argument structure is not a property of nouns per se. The noun event,
illustrated above, has no event/argument structure, even though, crudely speaking,
it has the right kind of meaning. Similarly, the noun trip: trips have beginnings
and ends, for example. Yet it is impossible to say *my trip for three weeks or *my
trip in three weeks (on the relevant reading) or *my frequent trip to the UK.3 This
leads us to conclude that the verb-like properties of complex event nouns are
attributable to the affixes which derive nouns from verbs, or to the verbal bases
themselves, or both. (Although the verb trip exists it has specialized meanings
which are semantically distant from the noun.)
2) As noted above, nouns which are identical in form to verbs do not
generally behave like complex event nominals, i.e. they are rigidly different from
verbs (recall offer, report above). Why? A simple-minded view suggests that they
should be most like verbs.
3) –ing nominals are always complex event nominals: Lebeaux (1986)
pointed out that they can take obligatory arguments, and Grimshaw (1990) and
Harley & Noyer (1998) offer further evidence.4
4) –(a)tion and –ment nominals are frequently ambiguous between
eventive and non eventive readings. See examination and replacement above. We
must conclude that nominal affixes, affixes which belong to a single syntactic
category (here nouns), can yield different interpretations, e.g. –ing vs. –ation.
Also, individual affixes can show a variety of behaviours, as –ment and –
(a)tion do.
3. Nouns and argument structure: where does responsibility lie?
The two models we are investigating attribute noun ~ argument structure
association to two different aspects of representation. In the structural model the
presence of argument structure follows from the presence of a VP node inside the
nominal structure (or perhaps some functional projection of VP). In the event
model the presence of argument structure follows from the existence of an event
structure, or “tier” matching the content of the nominal structure.
The event-based model posits a representation of the event structure of a
noun (or verb), which is linked to an argument structure. A complex event
nominal, by definition, denotes an event with an internal aspectual structure. For
3 Although my most frequent trip is grammatical, suggesting a different interpretation for frequent
here.
4 Apparent counterexamples seem to be arbitrary lexicalizations: a good living, hand-writing etc.
The references cited above discuss such cases.
4
Verbs, nouns and affixation
example, the noun replacement has the (obviously simplified) representations in
(5).
(5) a. replacement: the individual “z” in <x replaces y with z>
No aspect
b. replacement: the event <x replaces y with z>
Aspect – telic
In (5a), the noun corresponds to an argument of the verb, in (5b) it corresponds to
the event encoded by the verb: the noun is telic, like the base verb.
The structure-based model represents the difference between the two noun
types in terms of the presence of verbal functional layers and the height of
affixation. The higher the affix is in the structure, the more verbal properties the
derived noun will show. Complex event nominals are derived by high affixation.
In other words, complex event nominals contain some functional projection of
VP, while participant and simple event nominals lack such a projection. The main
idea behind the structural model could be described as follows: it is the syntactic
structure that gives rise to an event template which in turn determines the
interpretation of arguments (see Borer 2001). In other words, the event
interpretation arises through the presence of verbal functional layers in the
nominal structure.
4. Towards a description
Both of these theories offer the possibility of describing the facts as presented in
Section 2. The questions we are interested in arise in both the structural model and
the event model, and receive, surprisingly, answers of comparable status. This
becomes quite clear when we ask about explanation, rather than just description.
We move to this question in Section 5.
4.1 Preservation of argument structure under nominalization
We consider the generalizations in 1) – 4) above in turn.
1) Why is it only nouns which are derived from verbs that have complex event
readings/argument structure? This follows if only verbs, not nouns, can have
event structure/argument structure. Under this assumption, the only way for a
noun to be associated with an argument structure is for it to be derived from a
verb. This assumption must apparently be made by both theories.
In the event-based model, what must be eliminated is a representation of the form
shown in (6). (6) is intended to represent a noun with no internal verbal structure,
with an associated argument structure in angled brackets and an aspectual
structure in parentheses.
5
Alexiadou & Grimshaw
(6) [N ] <x,…. > (aspect: telic/atelic/…)
In contrast, (7) must be allowed:
(7) [V ] <x,…. > (aspect: telic/atelic/…)
We can hypothesize that the representation in (7) is the only one allowed for
verbs: all verbs must have an event structure and an argument structure. A noun
derived from a verb is represented this way, when the nominalization is event-
structure-preserving:
(8) [N [V ] .. ] <x,…. > (aspect: telic/atelic/…)
It is the verb, and not the noun, that is the source of the event structure and
argument structure, so the representations are consistent with the principles
governing verbs versus nouns. In this way, the model predicts that only nouns
derived from verbs can have argument structure.
It should be noted that the prediction is based on a stipulation (namely that
nouns never have event structure or argument structure) which presumably has
some more profound basis. In the structure-based theory, a virtually identical
stipulation gives an identical result.
If nouns cannot have argument structure, and verbs can or must, then a noun
which appears as the lexical head of an extended nominal can have no argument
structure. A noun which is built up from a V which is the lexical head of a VP
projection can have an argument structure, since it is contributed by the V.
The generalizations in 2), 3) and 4) all concern the effects of category-
changing (from V to N): by no affixation at all, by affixation with –ing, and by
affixation with –(a)tion or –ment.
As noted above, nouns which are identical in form to verbs do not generally
behave like complex event nominals. As we also noted above, the simplest theory
suggests that they should be most like verbs.
In the event based theory, the grammar stipulates which affixes “preserve
the verbal property of having argument structure”, which do not, and which do
both, perhaps because they are unspecified for this property. The suffix –ing is
argument-preserving, and –(a)tion and –ment are ambiguous/unspecified. Zero-
derivation involves an affix which is unpronounced, and this affix is opaque to
argument-transfer.5 This is in essence what is proposed in Grimshaw (1990: 67).
The cases of nominalization delineated above are thus represented as in (9-
11):
5 Referring to identical noun~verb pairs as “zero affixation” maximises the parallels between this
and the overt affixations.
6
Verbs, nouns and affixation
(9) “zero” affixation:
[N [V ] Ø ] <…> (aspect: ...)
(10) affixation with –ing:
[N [V ] –ing ] <x,….> (aspect: telic/atelic/…)
(11) affixation with –(a)tion; –ment:
a. [N [V ] –(a)tion] <…> (aspect: ...)
b. [N [V ] –(a)tion ] <x,….> (telic/atelic/…)
In a particular instantiation of the structural model it has been argued that
exponents such as –ing, –ation/–ment and zero morphology have rather distinct
specifications for insertions. Specifically, certain exponents, which are nominal-
category-determining heads can attach both to roots (“low/root attachment”)
and/or to some further layers of structure, “high/outer cycle attachment”, in
accordance with a low vs. high attachment parameter. (See Marantz 2001,
Alexiadou 2001, and Embick 2003 crucially echoing Abney's 1987 intuition.)
Some can attach only high or only low. In other words, all these exponents
express nominal categories, but they differ with respect to the height of affixation.
The zero affix discussed here, then, attaches only to roots:
(12) [D [n Ø [√
As a result, nominals derived from verbs by zero-affixation have no argument
structure, since they contain no verbal projections.
For gerundive –ing, the structural model posits a VoiceP, a vP and an AspP,
as projections above the root and below the DP, as in (13a-b). Here -ing is the
head of AspP in the case of verbal gerunds, and realises n in the case of nominal
gerunds, see Alexiadou & al. (2008), cf. Borer (2005):
(13) a. [D [ AspP ing [ VoiceP [ vP [ √
b. [D [ n ing [ VoiceP [ vP [ √
Since the root here is embedded within verbal functional projections, it preserves
its argument structure.
Finally, in both accounts, nominal exponents such as –(a)tion can be doubly
specified, or underspecified. In the structural model, as proposed in Alexiadou
(2001), this allows it to attach directly to stems/roots, giving rise to full nominals.
It may also attach to something larger, a VP (+functional projections). When
–(a)tion attaches to the root directly it gives rise to a nominal lacking argument
7
Alexiadou & Grimshaw
structure.6 When it attaches high, the result is a nominal that has some verbal
properties; in a sense, the root first becomes a verb and then a noun.7
(14) [D [ n –ation [ vP [ √
In other words, –(a)tion can appear in the structure in (14) as well as in the
structure in (12), while –ing can be ‘specified’ as entering only in the structure in
(13). Bare root nominals can appear only in the structure in (12). So particular
exponents of the nominal category, e.g. –(a)tion, –ing and the “zero” morpheme
appear in specific contexts (see Embick 2003). It will be obvious that
underspecification, or double specification in the event model, similarly has the
result that the –(a)tion and –ment suffixes can behave either like the null affix, or
like –ing.
This completes our summary of the alternative models of nominalization. Of
course, various combinations of these two approaches could be envisaged, but as
far as we can tell, the assessment of the proposals we give below would extend
equally to such mixed solutions.
4.2 The behaviour of adjectives and adverbs in nominalizations
A further consideration which potentially separates the event and structural
models is the discovery that complex event nominals can contain some adverbs.
This leads to the hypothesis that complex event nominals have at least VP inside
them, and possibly more verbal structure such as AspP (see Fu et al 2001, Hazout
1991, Alexiadou 2001).
The above authors note that certain adverbs are possible within complex
nominals (see (15) and (16) below).8
(15) a. The arrival of the trains promptly at the station ...
b. His careful destruction of the documents immediately ...
(16) *His explanation of the problem fortunately to the tenants ...
The logic of the argument is as follows: It is generally accepted that adverbs
modify verbal elements. But adverbs are distinguished (very roughly) into VP
modifiers, i.e. adverbs which modify only verbs/verb phrases (VP), e.g. The trains
arrived promptly at the station and sentence (S) modifiers, i.e. adverbs that
6 This analysis has been refined in Alexiadou (to appear) and Harley (to appear).
7 Note here that VoiceP is missing in -ation nominals. In agreement with Kratzer (1994), external
arguments are never assigned by the lexical entry, but by Voice. -Ation nominals lack Voice and
therefore they never have an external argument.
8 Siloni (1997) argues that apparent adverbials in the comparable Hebrew structures are really
adverbial PPs, adding another dimension to the interpretation of the adverb facts. See Borer (1993)
for counterarguments.
8
Verbs, nouns and affixation
modify propositions, e.g. Fortunately he explained the problem to the tenants.
Traditionally this distinction is resolved in terms of attachment of the modifier,
VP-adverbs attach to VPs, S-adverbs attach to sentences (TPs). The structural
hypothesis interprets the generalizations concerning the types of adverbs that can
be found within nominals as telling us something about the types of verbal
projections we can find, especially in view of recent typologies that recognize a
relationship between adverb types and inflectional material (see especially
Cinque's 1999 evidence for this richer picture of adverbs and also Alexiadou
1997). In this view, the admissibility of certain adverbs in complex event
nominals is not a fact that simply has to do with some kind of semantic
compatibility (a priori the semantics of a process nominal should not be different
from the semantics of a verb). Rather, it is a syntactic fact that has to do with the
principles that determine which elements can be attached at which positions in the
tree structure. This does not mean that the admissibility of adverbs has nothing to
do with the event interpretation associated with process nominals. But the
interpretation of such nominals as denoting events is not sufficient as an
explanation to the restrictions on the distribution of adverbs.
In contrast, the event-based hypothesis attributes the well-formedness of
adverbial modification directly to the event structure of complex event nominals
and asserts that the semantics of complex event nominals does distinguish them in
the relevant way from verbs (contra the position outlined in the previous
paragraph). Since these nominals denote events and not propositions, it is
expected that only event-related, and not proposition-related adverbials will be
able to appear, explaining (15) and (16) above.
In sum, adverbs modify semantic units, and they also appear in particular
configurational positions. Assuming that the semantic units and the configurations
match, the modified semantic units correspond to structural layers in the syntax.
So which licenses the adverbs? See Haider (2001), and Ernst (1998) for further
discussion.
5. Assessing the results
The success of these theories, or views, of nominalization can be judged by
comparison of what they stipulate, and what they derive from their premises. It is
striking, then, that both stipulate the same information, albeit in different form.
5.1 The stipulations
The “zero” morpheme is never transparent/always attaches to the root.
This is an accident in both models. The “zero” affix could always be transparent
to argument structure (its nominal forms thus always having an event structure
and argument structure). In the VP model, it could be attached only at the higher
level, and its nominal forms would then always have argument structure.
9
Alexiadou & Grimshaw
Moreover, the “zero” affix could instead be unspecified for transparency/level of
attachment, and thus behave like –ment and –(a)tion, in attaching to both low and
high levels of structure, or, to put it the other way, be like –ment and –(a)tion in
showing indifference to the complex event/simple event interpretations.
From a broader perspective, it is striking that nouns which are identical in
form to verbs do NOT behave like complex event nominals, since as pointed out
in Section 2 the simplest theory actually suggests that they should be most like
verbs. Both the models investigated here lack insight into this problem (cf. Borer
2005). In the event structure theory these nouns look the most like verbs, and
verbs have argument structure. Likewise, in the syntactic account, we would
expect zero nominal morphology to always attach high, since zero nouns most
resemble verbs.9
The –ing affix is always transparent/always attaches high up
Again, the –ing affix is merely stipulated to be only transparent/high attached,
when in principle it could be non-transparent/attached to the root, or unspecified.
Neither theory offers an explanation for the fact that this affix has to be
unspecified and the others may not be. One might want to speculate here that this
is related to the existence of the verbal suffix –ing, see Alexiadou & al. 2008,
Borer (2005).
The –(a)tion/–ment affixes are unspecified
Why is it these affixes that are unspecified? Why is it only these affixes that are
underspecified? Is it accidental that both of them are underspecified? Neither
theory answers these questions.10
Presumably there is more to the nominalization patterns than these theories
have been able to explicate. What is a surprise is the fact that what we might call
the articulation points are exactly the same. By this we mean that comparable
stipulations are necessary at comparable points in the structure of the theories.
The principal difficulties arise from the non-uniformity of deverbal
nominalization patterns: different affixation types exhibit different behaviour.
The successes and failures of the two models occur on exactly the same questions.
What one describes, the other describes. What one fails to explain, the other fails
to explain, and for fundamentally the same reasons.
9 Borer (to appear) assumes that zero morphology does not exist. In her analysis zero derived
nominals are simply lexical items inserted in nominal structure. Such nominals were never verbs,
and hence lack argument structure properties.
10
This is one of the reasons why other alternatives are pursued in Alexiadou (to appear) and
Harley (to appear). Both these papers argue that the difference between argument structure and
non-argument structure nominals does not depend on the presence of a verbal source. These
approaches attempt to derive the difference related to AS from the role of higher projections such
as Number.
10
Description:must conclude that nominal affixes, affixes which belong to a single syntactic category (here nouns), can yield different interpretations, e.g. –ing vs. –ation. Also, individual affixes can show a variety of behaviours, as –ment and –. (a)tion do. 3. Nouns and argument structure: where does
